High resolution micrographs are often of poor quality, due to various distortions and especially to a very low signal-to-noise ratio. Visual quality can be improved significantly by appropriately combining several observations of similar subjects. The aim of this project is to develop signal processing techniques to facilitate structure determination from such electron micrographs. Methods for translational and rotational alignment of different views of the same subject have been developed and implemented. Principal components and correspondence analysis have been used to seek important characteristics of the data. A procedure for automated statistical rejection of outlying samples has been proposed. These methods have been applied successfully to the structure determination of Herpes and T7 viruses by combining as many as 50 different observations of the same object.